What surprises me is that no-one has suggested the 'master/slave' model.
The master thermometer does not even need to be accurate, merely consistent, though accuracy is nice if you can get it. Mercury-in-glass lab thermometers (incliding the Kodak ones) are probably the most reliable; bimetallic dial drift most in my experience; and electronics may do almost anything without warning, no matter how reliable. Most don't, it's true, but they can.
All 'slaves' are calibrated to the 'master'. Let's say you have three thermometers, one 'master' and two 'slaves'. When the 'master' reads 20, one slave reads 19,5, the other, 20,5. For a standard 20, you therefore use the slaves accordingly.
Break a slave; replace it; recalibrate against the master. This one's 19.7. Fine. Still the same temperature...
My 'master' is a mercury-in-glass Brannan that agrees with another, identical, Brannan to 0.1 C. My 'slaves' vary widely for different purposes: spirit-in-glass, dial (for water baths), digital... Digital and dial are checked periodically; spirit-in-glass only when I acquire them.